Google DeepMind introduced Gemini 3.7 Flash on Aug. 13, 2026, positioning the release as an upgraded workhorse model designed for software engineering and automated agents. The launch arrived three weeks after the rollout of Gemini 3.6 Flash.
Google DeepMind set an introductory price through the end of the year at $0.75 per million input tokens and $3.75 per million output tokens. That structure cuts the original token pricing of Gemini 3.6 Flash in half.
Benchmark figures released by Google DeepMind showed accuracy increases across software development assessments. Gemini 3.7 Flash achieved 43.6 percent on FrontierCode 1.1 Main compared to 34.4 percent for 3.6 Flash, and reached 65.3 percent on DeepSWE v1.1 over 49.0 percent. The model also posted an Elo score of 1588 on Arena.ai's WebDev Arena, up from 1538.
Benchmark scores and demonstrations
Document understanding and workflow tests showed similar performance differences. Gemini 3.7 Flash recorded 34.0 percent on the GDP.pdf document evaluation, compared to 22.0 percent for 3.6 Flash. In AutomationBench, the model completed 30.4 percent of tasks compared to 17.0 percent previously.
Demonstrations published alongside the model paired Gemini 3.7 Flash with Nano Banana to generate real-time 3D game assets and with Gemini Omni to create parallax landing pages. The team also used the model in a three-agent graph loop to train a robotics system using multimodal understanding.
Platform availability and safety
Google added Gemini 3.7 Flash directly to Gemini Spark, its autonomous agent for Google AI Pro and Ultra subscribers across more than 160 countries. The integration links the model to Google Workspace tools to draft emails, aggregate files, and modify tracking documents.
Developers can access Gemini 3.7 Flash through Google Antigravity or the Gemini API in Google AI Studio and Android Studio, while enterprise teams can run it via the Gemini Enterprise Agent Platform. Google DeepMind also incorporated updated safety controls designed to prevent misuse involving cyber offense and chemical, biological, radiological, or nuclear materials.
